How much of Reinhart/Rogoff has survived?
April 19, 2013 7:59 am
by Gavyn Davies
The work of Carmen Reinhart and Ken Rogoff (RR) on public sector debt ratios, and their relationship with GDP growth, has been extraordinarily influential in academic and policy circles since 2010. Before this week, their statistical analysis, based on a 200-year database which they had painstakingly assembled covering dozens of countries, had appeared to establish an important stylised fact: that debt ratios above 90 per cent were associated with much lower rates of GDP growth than debt ratios under 90 per cent. The sudden drop in growth at a debt ratio similar to that reached in many developed economies acted as a wake up call to governments and encouraged the adoption of austerity programmes.
This week, a paper by Thomas Herndon, Michael Ash and Robert Pollin (HAP) argued that the RR stylised fact was based on simple statistical errors, including a spreadsheet error which RR have now acknowledged. Their critique of the original RR stylised fact promises to establish an alternative conventional wisdom, which is that high public debt ratios are never damaging for GDP growth. But the truth is more complicated than that, and far less certain.
For those who have not been paying attention to this dispute, here is my (hopefully not too biased) commentary on the current state of the debate.
First, the original work of RR never really established that GDP growth was likely to decline suddenly once a 90 per cent debt threshold had been exceeded.
In the many results published by RR in several papers, their numbers tended to demonstrate that there was a significant decline in growth at that point, but no dramatic discontinuity. A typical result is shown in the second column of the table, based on median growth rates, which suggests that GDP growth falls by a little over 1 per cent when the 90 per cent debt ratio is exceeded.
However, in one particular calculation, they did identify a large and sudden decline in average growth, as shown in the pink box in the table. This was based on a controversial method of calculating average or mean growth rates from 1945-2009, and that one result was sufficiently eye-catching to gain a great deal of attention. HAP say the key result is simply based on error. They say the “correct” figure is 2.2 per cent, not -0.1 per cent.
Second, not all of the difference between RR and HAP is attributable to the spreadsheet error which has become infamous this week. In fact, it appears that only 0.3 per cent stems from this, with another 0.1 per cent coming from a transcription mistake. The remainder of the difference is due to a combination of missing data in the original RR paper, and to a methodological dispute about how countries should be weighted to produce the final results.
My interpretation is that HAP have had the better of this element in the debate, because the RR method has the very odd effect that a few years of New Zealand data in the late 1940s appear to have depressed their critical estimate of the overall growth effect by over 1.5 per cent, which surely cannot make much sense. This alone explains most of the difference between HAP and RR.
Third, although the most dramatic version of the RR stylised fact has not survived, even HAP confirm that the RR cross-country database does show that higher public debt ratios are usually associated with lower GDP growth rates. This is a standard result from macro-economics, which has been established over long periods in many other studies. No surprise there.
Fourth, as Paul Krugman has argued many times, this negative correlation between debt and growth does not establish the direction of causation. RR say they never claimed this, but many others certainly inferred it from their work. It is possible that high debt ratios cause reduced growth, through confidence and interest rate effects, but it is also possible that high debt ratios cause reduced growth, through confidence and interest rate effects, but it is also possible that low growth reduces tax receipts and therefore causes high debt ratios. It is not one way traffic.
Fifth, the work which has been done by Arindrajit Dube on the timing of these effects suggests that changes in growth precede changes in debt, and not the other way around. On the surface, this seems to be supportive of a Keynesian effect, in which a recession takes hold (eg because of a financial crash) and public debt subsequently rises. But even that is not entirely clear.
Imagine a situation in which public expenditure is running out of control, so that households fear a collapse of market confidence, or a rise in tax rates, at some uncertain point in the future. They might then cut their spending today, in which case GDP growth will decline before debt rises; but the true cause of the problem will be too much public spending, not too little. In other words, post hoc ergo propter hoc does not always apply.
Sixth, there really is no reason why we should expect the relationship between public debt and GDP growth to be stable through time and across all countries. In a fully employed economy (which applies to most countries in the period under consideration above), a deliberate rise in the budget deficit will raise interest rates, and crowd out private investment. This will reduce long term GDP growth rates.
However, if the economy is working well below capacity, a rise in the budget deficit may not raise interest rates, but may instead raise aggregate demand and thus boost GDP growth. Under some circumstances, this might even reduce the debt ratio for a while.
In summary, the most dramatic version of the RR stylised fact is no longer a stylised fact. RR were right to argue that, over most normal periods, higher public debt has been associated with lower real GDP growth rates, but a sudden discontinuity at 90 per cent is not proven. Furthermore, causation might work in both directions, depending on economic circumstances. The timing of these effects is not a definitive indicator of true causation, and the relationship may be very different in a time of full employment from a time of high unemployment.
The moral of this story is that it is an illusion to expect that the complicated relationship between public debt and GDP growth will always and everywhere be the same.
This week, a paper by Thomas Herndon, Michael Ash and Robert Pollin (HAP) argued that the RR stylised fact was based on simple statistical errors, including a spreadsheet error which RR have now acknowledged. Their critique of the original RR stylised fact promises to establish an alternative conventional wisdom, which is that high public debt ratios are never damaging for GDP growth. But the truth is more complicated than that, and far less certain.
For those who have not been paying attention to this dispute, here is my (hopefully not too biased) commentary on the current state of the debate.
First, the original work of RR never really established that GDP growth was likely to decline suddenly once a 90 per cent debt threshold had been exceeded.
In the many results published by RR in several papers, their numbers tended to demonstrate that there was a significant decline in growth at that point, but no dramatic discontinuity. A typical result is shown in the second column of the table, based on median growth rates, which suggests that GDP growth falls by a little over 1 per cent when the 90 per cent debt ratio is exceeded.
However, in one particular calculation, they did identify a large and sudden decline in average growth, as shown in the pink box in the table. This was based on a controversial method of calculating average or mean growth rates from 1945-2009, and that one result was sufficiently eye-catching to gain a great deal of attention. HAP say the key result is simply based on error. They say the “correct” figure is 2.2 per cent, not -0.1 per cent.
Second, not all of the difference between RR and HAP is attributable to the spreadsheet error which has become infamous this week. In fact, it appears that only 0.3 per cent stems from this, with another 0.1 per cent coming from a transcription mistake. The remainder of the difference is due to a combination of missing data in the original RR paper, and to a methodological dispute about how countries should be weighted to produce the final results.
My interpretation is that HAP have had the better of this element in the debate, because the RR method has the very odd effect that a few years of New Zealand data in the late 1940s appear to have depressed their critical estimate of the overall growth effect by over 1.5 per cent, which surely cannot make much sense. This alone explains most of the difference between HAP and RR.
Third, although the most dramatic version of the RR stylised fact has not survived, even HAP confirm that the RR cross-country database does show that higher public debt ratios are usually associated with lower GDP growth rates. This is a standard result from macro-economics, which has been established over long periods in many other studies. No surprise there.
Fourth, as Paul Krugman has argued many times, this negative correlation between debt and growth does not establish the direction of causation. RR say they never claimed this, but many others certainly inferred it from their work. It is possible that high debt ratios cause reduced growth, through confidence and interest rate effects, but it is also possible that high debt ratios cause reduced growth, through confidence and interest rate effects, but it is also possible that low growth reduces tax receipts and therefore causes high debt ratios. It is not one way traffic.
Fifth, the work which has been done by Arindrajit Dube on the timing of these effects suggests that changes in growth precede changes in debt, and not the other way around. On the surface, this seems to be supportive of a Keynesian effect, in which a recession takes hold (eg because of a financial crash) and public debt subsequently rises. But even that is not entirely clear.
Imagine a situation in which public expenditure is running out of control, so that households fear a collapse of market confidence, or a rise in tax rates, at some uncertain point in the future. They might then cut their spending today, in which case GDP growth will decline before debt rises; but the true cause of the problem will be too much public spending, not too little. In other words, post hoc ergo propter hoc does not always apply.
Sixth, there really is no reason why we should expect the relationship between public debt and GDP growth to be stable through time and across all countries. In a fully employed economy (which applies to most countries in the period under consideration above), a deliberate rise in the budget deficit will raise interest rates, and crowd out private investment. This will reduce long term GDP growth rates.
However, if the economy is working well below capacity, a rise in the budget deficit may not raise interest rates, but may instead raise aggregate demand and thus boost GDP growth. Under some circumstances, this might even reduce the debt ratio for a while.
In summary, the most dramatic version of the RR stylised fact is no longer a stylised fact. RR were right to argue that, over most normal periods, higher public debt has been associated with lower real GDP growth rates, but a sudden discontinuity at 90 per cent is not proven. Furthermore, causation might work in both directions, depending on economic circumstances. The timing of these effects is not a definitive indicator of true causation, and the relationship may be very different in a time of full employment from a time of high unemployment.
The moral of this story is that it is an illusion to expect that the complicated relationship between public debt and GDP growth will always and everywhere be the same.
0 comments:
Publicar un comentario